Feature Selection Methodology Research Articles

Feature selection on real-time application using hybrid of flamingo search algorithm and improved non-dominated sorting genetic algorithm III

ABSTRACT Feature selection plays an essential role in enhancing the efficiency and effectiveness of machine learning models, particularly in the context of large-scale datasets where the dimensionality of features presents significant challenges. This research presents a novel approach to feature selection, a critical aspect of enhancing machine learning model efficiency, particularly in the context of large-scale datasets. By integrating the Flamingo Search Algorithm (FSA), Non-dominated Sorting Genetic Algorithm III (NSGA-III), and Regularised Extreme Learning Machine (RELM), the proposed method addresses limitations in existing feature selection and multi-objective optimisation algorithms. Leveraging FSA’s emulation of flamingo behaviours, the approach achieves a balance between global exploration and local exploitation, mitigating issues like premature convergence and local optima. Integration with NSGA-III enhances multi-objective optimisation capabilities, maintaining a delicate equilibrium between convergence and diversity. FSA-RELM is employed for accurate feature assessment, given its rapid learning and suitability for large datasets with multiple labels. Experimental evaluations demonstrate the proposed method’s superiority in feature selection accuracy, classification performance, and computational efficiency compared to existing approaches. This research contributes to advancing feature selection methodologies, offering a comprehensive solution for high-dimensional datasets in machine learning and data mining applications.

FSSDroid: Feature subset selection for Android malware detection

Android malware has become an increasingly important threat to individuals, organizations, and society, posing significant risks to data security, privacy, and infrastructure. As malware evolves in sophistication and complexity, the detection and mitigation of these malicious software instances have become more challenging and time consuming since the required number of features to identify potential malware can be very high. To address this issue, we have developed an effective feature selection methodology for malware detection in Android. The critical concern in the field of malware detection is the complexity of algorithms and the use of features that are used to detect malware. The present paper delivers a methodology for pre-processing datasets to select the most optimal features that will allow detecting malware, while maintaining very high accuracy. The proposed methodology has been tested on two real world datasets and the results indicate that the number of features is significantly reduced from 489 to between 19 and 28 for the first dataset and from 9503 to between 9 and 27 for the second dataset, whilst the accuracy is maintained as if all features were used.

Ensemble Feature Selection for Clustering Damage Modes in Carbon Fiber‐Reinforced Polymer Sandwich Composites Using Acoustic Emission

Acoustic emission (AE) serves as a noninvasive technique for real‐time structural health monitoring, capturing the stress waves produced by the formation and growth of cracks within a material. This study presents a novel ensemble feature selection methodology to rank features highly relevant with damage modes in AE signals gathered from edgewise compression tests on honeycomb‐core carbon fiber‐reinforced polymer. Two distinct features, amplitude and peak frequency, are selected for labeling the AE signals. An ensemble‐supervised feature selection method ranks feature importance according to these labels. Using the ranking list, unsupervised clustering models are then applied to identify damage modes. The comparative results reveal a robust correlation between the damage modes and the features of counts and energy when amplitude is selected. Similarly, when peak frequency is chosen, a significant association is observed between the damage modes and the features of partial powers 1 and 2. These findings demonstrate that, in addition to the commonly used features, other features, such as partial powers, exhibit a correlation with damage modes.

A new adaptive elastic loss for robust unsupervised feature selection

The sparse-based unsupervised dimension reduction technique has been demonstrated to be an effective tool for dealing with high-dimensional data. Among them, the sparse regularization technique is used to introduce the prior knowledge of the model and helps unsupervised dimension reduction algorithms to complete the feature selection. However, the sparse regularization only considers the sparsity of representation. It ignores the group effect of variables among which the pairwise correlations are very high. To solve these issues, we present a novel regularization approach based on adaptive elastic loss in this study. Then, utilizing the regularization method, an efficient unsupervised feature selection methodology is given. Additionally, we devise an efficient alternative optimization algorithm to resolve the complex optimization issue of our proposed method and conduct a theoretical analysis of its computational complexity and convergence. Eventually, we conduct comprehensive experiments to test the validity of our proposed method, and real-world data and a simulation study demonstrate that our proposed method outperforms many state-of-the-art unsupervised feature selection methods in terms of clustering and classification.

#328 Non-invasive assessment of urinary exfoliated proximal tubule cells using multispectral autofluorescence imaging features for early detection of CKD

Abstract Background and Aims Early detection of CKD is important to allow for timely intervention which may reduce adverse clinical outcomes and cost of renal replacement therapies. Histopathological diagnosis of CKD remains the gold standard, but routine kidney biopsy is costly and associated with risks such as pain and bleeding. There is a need to develop accurate non-invasive methods for early diagnosis and prognostication of CKD. Kidney cell exfoliation into urine is an active process, which may generate significant information regarding an individual's kidney status. In particular, proximal tubule cells (PTCs) make up majority of the kidney mass and represents the hallmark of CKD in reflecting the degree of tubulointerstitial fibrosis and atrophy. Multispectral assessment of native cell autofluorescence has been specifically shown to be sensitive to metabolic changes and oxidative stress, which are key factors involved in kidney function decline. Hence, we evaluated whether multispectral autofluorescence signals in urinary exfoliated PTCs reflect kidney function as determined by eGFR. Method Individuals aged 18 or above were included and informed consent was obtained from all participants involved in this study. Spot urine samples were collected and stored at -80°C. On assessment, exfoliated PTCs were extracted from thawed urine using a validated specific immunomagnetic separation method based on anti-CD13 and anti-SGLT2 antibodies. Multispectral autofluorescence imaging was performed using a custom-made autofluorescence microscopy system (standard Olympus iX83™ fluorescence microscope and multispectral excitation lamp from Quantitative™, AU). Images were captured by a Nüvü™ EMCCD camera HNü 1024 cooled to -65°C in our laboratory to reduce sensor-induced noise. For the 34 image channels taken per sample, image acquisition times of up to 5 seconds per channel, with multiple averaging (3 times) was applied to optimize image quality. Image analysis was conducted through a workflow encompassing data preprocessing, data normalization and outlier handling, followed by basic model training and evaluation. An entropy-based image feature selection methodology was initially completed by hand, with results subsequently compared to that generated from the AutoGluon machine-learning framework [1]. The models developed from this process were assessed using ROC (Receiver Operating Characteristic) curves and AUC (Area Under the Curve) values. Two discriminative studies were performed – an overall study between exfoliated PTCs of individuals eGFR≥60 and eGFR < 60 ml/min/1.73 m2 and a subgroup study between that of individuals eGFR≥90 and eGFR 60-90 ml/min/1.73 m2. Results 40 individuals were included in our study. 20 individuals had eGFR≥60 and 20 had eGFR < 60 ml/min/1.73 m2. Multispectral assessment of cell autofluorescence features differentiated between exfoliated PTCs of these 2 groups with AUC value of 0.97 (95% CI 0.94-1.00) (Fig. 1), model accuracy being 82%. Subgroup analysis included 20 individuals of which 10 had eGFR≥90 and 10 had eGFR 60-90 ml/min/1.73 m2. Multispectral assessment of cell autofluorescence features differentiated between exfoliated PTCs of these 2 groups with AUC value of 0.99 (95% CI 0.99-1.00) (Fig. 2), model accuracy being 85%. Conclusion An excellent degree of differentiation between urinary PTCs derived from those with normal or high eGFR and mild degrees of CKD is now reliably demonstrated, complementing conclusions from our group's original small pilot study which evaluated between urinary PTCs from CKD and non-CKD cohorts [2]. Further studies are required to determine whether our novel method can be used to non-invasively predict individuals at risk of progressive kidney disease and to monitor whether intervention is effective.

Acute lymphocytic leukemia detection and subtype classification via extended wavelet pooling based-CNNs and statistical-texture features

Acute lymphoblastic leukemia (ALL) is considered the most fatal form of leukemia (also known as blood cancer). It propagates quickly among adults and children and could lead to their death. Early detection of ALL and ALL subtypes is the key factor in selecting effective treatment types and improving survival rates. However, routine diagnostic approaches have several drawbacks. Computer-assisted diagnosis (CAD) is the perfect solution to avoid these challenges and achieve a fast and accurate diagnosis. Current CAD models require enhancement/segmentation processing. Besides, they are either dependent on deep learning (DL) models or handcrafted features along with machine learning. Those CADs that employed DL approaches relied solely on spatial information during the training procedure. However, learning them with spectral temporal and temporal representations could improve performance. Furthermore, integrating deep features from DL models along with handcrafted features can increase the discrimination ability of attributes in medical image classification. This study aims to propose a novel CAD for ALL detection and subtype classification without pre-segmentation or enhancement steps. The proposed CAD extends the conventional DL models of convolutional neural networks by introducing an additional wavelet pooling, accompanied by a dense layer or a long-short-term memory (LSTM) layer, and then a SoftMax layer, acquiring spectral-temporal information along with temporal information. To further improve the framework's ability to discriminate, the introduced CAD then combines the wavelet-based deep features of every CNN with numerous handcrafted attributes. Afterward, a feature selection methodology is utilized to create a model with limited features and improved accuracy. The performance results show that the novel CAD is capable of achieving 100% ALL detection accuracy, as well as 100% ALL-subtype classification accuracy with just 88 and 146 features. Thus, this CAD can be employed to assist pathologists in the rapid and precise ALL identification and subcategories recognition.

Enhancing gastric cancer early detection: A multi-verse optimized feature selection model with crossover-information feedback

Gastric cancer (GC), an acknowledged malignant neoplasm, threatens life and digestive system functionality if not detected and addressed promptly in its nascent stages. The indispensability of early detection for GC to augment treatment efficacy and survival prospects forms the crux of this investigation. Our study introduces an innovative wrapper-based feature selection methodology, referred to as bCIFMVO-FKNN-FS, which integrates a crossover-information feedback multi-verse optimizer (CIFMVO) with the fuzzy k-nearest neighbors (FKNN) classifier. The primary goal of this initiative is to develop an advanced screening model designed to accelerate the identification of patients with early-stage GC. Initially, the capability of CIFMVO is validated through its application to the IEEE CEC benchmark functions, during which its optimization efficiency is measured against eleven cutting-edge algorithms across various dimensionalities—10, 30, 50, and 100. Subsequent application of the bCIFMVO-FKNN-FS model to the clinical data of 1632 individuals from Wenzhou Central Hospital—diagnosed with either early-stage GC or chronic gastritis—demonstrates the model's formidable predictive accuracy (83.395%) and sensitivity (87.538%). Concurrently, this investigation delineates age, gender, serum gastrin-17, serum pepsinogen I, and the serum pepsinogen I to serum pepsinogen II ratio as parameters significantly associated with early-stage GC. These insights not only validate the efficacy of our proposed model in the early screening of GC but also contribute substantively to the corpus of knowledge facilitating early diagnosis.

Using Feature Selection Enhancement to Evaluate Attack Detection in the Internet of Things Environment

The rapid evolution of technology has given rise to a connected world where billions of devices interact seamlessly, forming what is known as the Internet of Things (IoT). While the IoT offers incredible convenience and efficiency, it presents a significant challenge to cybersecurity and is characterized by various power, capacity, and computational process limitations. Machine learning techniques, particularly those encompassing supervised classification techniques, offer a systematic approach to training models using labeled datasets. These techniques enable intrusion detection systems (IDSs) to discern patterns indicative of potential attacks amidst the vast amounts of IoT data. Our investigation delves into various aspects of supervised classification, including feature selection, model training, and evaluation methodologies, to comprehensively evaluate their impact on attack detection effectiveness. The key features selected to improve IDS efficiency and reduce dataset size, thereby decreasing the time required for attack detection, are drawn from the extensive network dataset. This paper introduces an enhanced feature selection method designed to reduce the computational overhead on IoT resources while simultaneously strengthening intrusion detection capabilities within the IoT environment. The experimental results based on the InSDN dataset demonstrate that our proposed methodology achieves the highest accuracy with the fewest number of features and has a low computational cost. Specifically, we attain a 99.99% accuracy with 11 features and a computational time of 0.8599 s.

A novel two-stage wrapper feature selection approach based on greedy search for text sentiment classification

Sentiment analysis is a crucial step in obtaining subjective data from online text sources. Nevertheless, the substantial challenge of high dimensionality prevails within text classification. Addressing this, dimension reduction emerges as a valuable approach to enhance the efficacy of classification in the domain of machine learning. The discerning removal of redundant features not only expedites training processes but also bolsters the achievement of accurate classifications. It is worth noting that the effectiveness of distinct feature selection methodologies can be contingent upon the unique attributes inherent in diverse datasets. Within the purview of this investigation, a novel two-stage approach is introduced, characterized by a greedy search-based wrapper feature selection algorithm. The underpinning of this algorithm involves leveraging the outcomes yielded by filter-based feature selection techniques to establish a prioritized sequence for the scrutiny of features within the proposed framework. This strategic sequencing harnesses the cumulative insights from a series of filter-based methodologies, thereby facilitating the curation of feature subsets that underscore pivotal attributes. Nonetheless, it is acknowledged that the greedy selection approach primarily favors features with high-ranking scores, and thus, it may not adequately evaluate the potential of feature combinations that involve low-scoring elements. An extensive experimental inquiry was conducted across widely recognized sentiment analysis datasets to assess the performance of the introduced methodology. The computational findings eloquently demonstrate that the proposed algorithm attains an average accuracy of 96.88% for nine public sentiment datasets and 94.43% for the Humir datasets when coupled with the multinomial Naive Bayes classifier. Furthermore, the experimental outcomes conspicuously establish the superiority of the proposed technique in state-of-the-art studies across the same set of nine sentiment datasets and the Humir datasets.

Phishing URL Detection using Machine Learning

Abstract: Phishing attacks pose a significant threat in cyberspace, Cybercrimes such as Phishing means accessing personal data and violating security through the Internet and its main aim is to steal the information from the users using different techniques in that the primary one is Phishing which demands effective detection mechanisms. This study evaluates the performance of Gradient Boosting Classifier, Random Forest, and Decision Tree machine learning models in conjunction with feature selection techniques namely SelectKBest and Chi-Square. Initially, a comprehensive feature set of 30 attributes achieved a baseline accuracy of 97.4%. Through SelectKBest and Chi-Square feature selection methods, 13 key features were identified, leading to a slightly decreased accuracy of 95.6% upon model retraining. This research highlights the importance of feature selection in enhancing phishing detection accuracy while maintaining model interpretability. Technologies such as NumPy, Pandas, Matplotlib, Scikit-learn, and Flask drive. The project, emphasising the exploration of ML models, EDA (Exploratory Data Analysis) on phishing data, and understanding feature importance. The Machine Learning Models like Gradient Boosting Classifier, Random Forest and Decision tree are used to detect whether the given URL is Malicious URL or Legitimate URL using Comprehensive Feature set and Key Feature set. This Models and Feature sets are compared based on the Performance Metrics namely Accuracy, Precision, Recall and F1- score. This study contributes to advancing cyber defence mechanism through the fusion of sophisticated Machine Learning algorithms and meticulous Feature Selection methodologies.

A comprehensive view of biometric payment in retailing: A complete study from user to expert

The new financial and commercial scenario driven by technological advances has undergone a rapid reconfiguration in the recent years. Innovation has generated new payment alternatives that are transforming the concept of money and payment habits among consumers. One of the most novel payment systems nowadays is known as biometric payment. Improved payment systems will improve retailing and consumer services. The aim of this study is to develop an analysis of biometric payment based on two complementary studies. In the first one, the variables predicting the intention to use this technology are determined on a sample of 1905 potential users by means of different feature selection methodologies from artificial intelligence in a holistic model that integrates the principles of the UTAUT2 model, the General Risk Theory, and the Trust Theory. In the second study, two panels of Fintech industry experts compare these results. The overall insights obtained show that perceived risk, trust, and social influence are the variables that, from the experts’ experience, users consider most important when employing this technology. This research provides useful information for financial and business decision-makers in companies interested in commercializing this type of technology.

A Comprehensive Feature Engineering Approach for Breast Cancer Dataset

Breast cancer continues to pose a significant challenge in the field of healthcare, serving as the primary cause of cancer-related deaths in women on a global scale. The present study aims to investigate the intricate relationship between breast cancer, statistical analysis, and feature engineering. By conducting an extensive analysis of a comprehensive dataset and employing sophisticated statistical methodologies, this research endeavor aims to unveil concealed insights that can enrich the medical community's existing knowledge base. Through the implementation of rigorous feature selection and extraction methodologies, the overarching aim is to augment the comprehension of breast cancer. Moreover, the study showcases the successful incorporation of univariate and bivariate analysis in order to enhance the accuracy of diagnostic procedures. The convergence of these disciplines exhibits considerable promise in the realm of breast cancer detection and prediction, facilitating cooperative endeavours aimed at addressing this widespread malignancy.

Gait disorder classification based on effective feature selection and unsupervised methodology

In gait stability analysis, patients suffering from dysfunction problems are impacted by shifts in their dynamic balance. Monitoring the patients’ progress is important for allowing physicians and patients to observe the rehabilitation process accurately. In this study, we designed a new methodology for classifying gait disorders to quantify patients’ progress. The dataset in this study includes 84 measurements of 37 patients based on a physician’s opinion. In this study, the system, which includes a Kinect camera to observe and store the frames of patients walking down a hallway, a key-point detector to detect the skeletal key points, and an encoder transformer classifier network integrated with generator–discriminator networks (ET-GD), is designed to evaluate the classification of gait dysfunction. The detector extracts the skeletal key points of patients. After feature engineering, the selected high-level features are fed into the proposed neural network to analyse patient movement and perform the final evaluation of gait dysfunction. The proposed network is inspired by the 1D encoder transformer, which is integrated with two main networks: a network for classification and a network to generate fake output data similar to the input data. Furthermore, we used a discriminator structure to distinguish between the actual data (input) and fake data (generated data). Due to the multi-structural networks in the proposed method, multi-loss functions need to be optimised; this increases the accuracy of the encoder transformer classifier.

Inovativna primena veštačkih neuronskih mreža za efikasnu lokalizaciju prslina rotacionog vratila

Rotational shafts are pivotal components in industrial settings and are responsible for transmitting torque and rotational motion. Despite their significance, these shafts are susceptible to faults, particularly cracks, which can adversely affect the system's performance and safety. Hence, efficient crack detection and diagnosis ensure safety, reliability, and costeffectiveness. This research aims to develop an Artificial Neural Network (ANN) model that can effectively identify cracks occurring at different depths and locations in rotating shafts, which operate at varying rotational speeds. Vibration signals were obtained and subjected to preprocessing using a bandpass filter to isolate the shaft signals from other components. Subsequently, time-domain statistical features were extracted from the filtered signals. An optimal feature selection methodology was employed to rank the extracted features, and the highest-ranking features were chosen for training the ANN model. The findings of this research indicate that the developed model achieved a classification accuracy of 94.4%.

An Auto-encoded Warm Equilibrium Automated Learning (AE2L) Model for Automatic Recognition and Classification of Autism Spectrum Disorder

Autism spectrum disorder (ASD) is a neurological condition characterized by difficulties with communication and socializing, and repetitive activities. If the underlying reason is hereditary, early detection is still important, and machine learning offers a fascinating way to identify the condition more rapidly and economically. However, the unique issues of higher computational costs, longer execution times, and lower effectiveness affect the traditional methods. The proposed project aims to create an automated artificial intelligence tool for ASD identification that combines several state-of-the-art mining techniques to deliver the best possible level of disease prediction accuracy. For accurate and effective ASD identification, this research suggests an automated and lightweight method dubbed the auto-encoded warm equilibrium automated learner. To speed up the handicap detection process, a unique warm optimized feature selection methodology is applied to minimize the dimensionality of attributes. In addition, auto-encoded term memory equilibrium learning, a powerful deep learning technique, is designed to accurately and less frequently detect ASD from the given data. Moreover, the classifier performs better when hyperparameters are tuned using the equilibrium optimization model. The results of the proposed AE2L model have been tested and validated using a variety of parameters utilizing the well-known ASD dataset that was taken from the UCI repository.

Unveiling the unborn: Advancing fetal health classification through machine learning

Fetal health classification is a critical task in obstetrics, which enables early identification and management of potential health problems. However, it remains a challenging task due to data complexity and limited labeled samples. This research paper presents a novel machine-learning approach for fetal health classification, leveraging a LightGBM classifier trained on a comprehensive dataset. The proposed model achieves an impressive accuracy of 98.31% on a test set. The findings demonstrate machine learning can potentially enhance fetal health classification, offering a more objective and accurate assessment. Notably, the presented approach combines various features, such as fetal heart rate, uterine contractions, and maternal blood pressure, to provide a comprehensive evaluation. This methodology holds promise for improving early detection and treatment of fetal health issues, ensuring better outcomes for both mothers and babies. In addition to the high accuracy, the novelty of this approach lies in its comprehensive feature selection and assessment methodology. By incorporating multiple data points, this model offers a more holistic and reliable evaluation compared to traditional methods. This research has significant implications in the field of obstetrics, paving the way for advancements in early detection and intervention of fetal health concerns. Future work involves validating the model on a larger dataset and developing a clinical application. Ultimately, we anticipate that our research will revolutionize the assessment and management of fetal health, contributing to improved healthcare outcomes for expectant mothers and their fetuses.

Machine learning approach for Migraine Aura Complexity Score prediction based on magnetic resonance imaging data

BackgroundPrevious studies have developed the Migraine Aura Complexity Score (MACS) system. MACS shows great potential in studying the complexity of migraine with aura (MwA) pathophysiology especially when implemented in neuroimaging studies. The use of sophisticated machine learning (ML) algorithms, together with deep profiling of MwA, could bring new knowledge in this field. We aimed to test several ML algorithms to study the potential of structural cortical features for predicting the MACS and therefore gain a better insight into MwA pathophysiology.MethodsThe data set used in this research consists of 340 MRI features collected from 40 MwA patients. Average MACS score was obtained for each subject. Feature selection for ML models was performed using several approaches, including a correlation test and a wrapper feature selection methodology. Regression was performed with the Support Vector Machine (SVM), Linear Regression, and Radial Basis Function network.ResultsSVM achieved a 0.89 coefficient of determination score with a wrapper feature selection. The results suggest a set of cortical features, located mostly in the parietal and temporal lobes, that show changes in MwA patients depending on aura complexity.ConclusionsThe SVM algorithm demonstrated the best potential in average MACS prediction when using a wrapper feature selection methodology. The proposed method achieved promising results in determining MwA complexity, which can provide a basis for future MwA studies and the development of MwA diagnosis and treatment.

Intelligent Fault Diagnosis of Rolling Bearings Based on a Complete Frequency Range Feature Extraction and Combined Feature Selection Methodology

The utilization of multiscale entropy methods to characterize vibration signals has proven to be promising in intelligent diagnosis of mechanical equipment. However, in the current multiscale entropy methods, only the information in the low-frequency range is utilized and the information in the high-frequency range is discarded. In order to take full advantage of the information, in this paper, a fault feature extraction method utilizing the bidirectional composite coarse-graining process with fuzzy dispersion entropy is proposed. To avoid the redundancy of the full frequency range feature information, the Random Forest algorithm combined with the Maximum Relevance Minimum Redundancy algorithm is applied to feature selection. Together with the K-nearest neighbor classifier, a rolling bearing intelligent diagnosis framework is constructed. The effectiveness of the proposed framework is evaluated by a numerical simulation and two experimental examples. The validation results demonstrate that the extracted features by the proposed method are highly sensitive to the bearing health conditions compared with hierarchical fuzzy dispersion entropy, composite multiscale fuzzy dispersion entropy, multiscale fuzzy dispersion entropy, multiscale dispersion entropy, multiscale permutation entropy, and multiscale sample entropy. In addition, the proposed method is able to identify the fault categories and health states of rolling bearings simultaneously. The proposed damage detection methodology provides a new and better framework for intelligent fault diagnosis of rolling bearings in rotating machinery.

Adaptive fuzzy multi-neighborhood feature selection with hybrid sampling and its application for class-imbalanced data

For imbalanced data, classification efficiency degrades significantly due to the missing information for the positive class, and existing sampling schemes do not consider the distributions of samples. Additionally, the global parameters of fuzzy neighborhoods are set manually. These defects affect the effectiveness of classifier. To address these problems, we offer an adaptive fuzzy multi-neighborhood feature selection methodology with intercluster distance-based hybrid sampling for class-imbalanced data. First, the number of clusters can be defined in terms of the number of samples in the negative or positive class. The initial centers of the clusters are determined according to the number of clusters, and the dissimilarity and similarity measures are calculated by using the intercluster distances between samples. Then, the cluster center, fuzzy membership matrix, and intercluster distance are studied, and then the optimization objective function is designed. The hybrid sampling scheme can be used to combine the generated positive class samples and negative class samples and obtain a class-balanced system. Second, according to the sample distribution, the standard deviation and a set of adaptive fuzzy multi-neighborhood radii are designed. A fuzzy multi-neighborhood similarity relation is defined by introducing a Gaussian kernel model to obtain a fuzzy multi-neighborhood granule, and an improved fuzzy multi-neighborhood rough set model is provided. Uncertain measures of fuzzy neighborhood systems are evaluated by the positive region and dependency. Third, by integrating fuzzy dependence with fuzzy complementary condition entropy, fuzzy multi-neighborhood complementary mutual information is provided on two viewpoints of algebra and information. Finally, a heuristic feature subset selection methodology for imbalanced classification with hybrid sampling using fuzzy c-means clustering is studied to obtain this excellent set of features. Experiments on 26 imbalanced datasets show the effectiveness of our designed algorithm.

Explainable AI-driven wavelength selection for hyperspectral imaging of honey products

Hyperspectral imaging (HSI) offers a rapid and non-destructive method for inspecting various food and agricultural products. Though HSI delivers rich data with its continuous spectral bands, it simultaneously presents challenges due to its high dimensionality, potential redundancy, and noise interference. To tackle these challenges, dimension reduction becomes essential in HSI data processing. Our study integrated 11 explainable artificial intelligence (XAI) algorithms with convolutional neural networks (CNNs) to craft an advanced wavelength selection method tailored for honey product analysis. We developed two CNN models and applied five normalization techniques to differentiate honey derived from 23 distinct botanical sources. Through the XAI algorithms, we determined the significance of each spectral wavelength, pinpointing the most characteristic bands. We then used the top eight spectral bands—selected from the averaged significance scores across the XAI algorithms—to optimize the combination of CNN classifiers and normalization techniques. This approach not only streamlines HSI data analysis but also validates the robustness of our XAI-based feature selection methodology. Experimental results demonstrated that the CNN achieved a macro average F1 score of ≥0.99 using just 8 spectral bands selected by averaging the feature importance scores from different XAI algorithms. The generalization ability of our proposed band selection method was further demonstrated by experiments on three HSI remote sensing databases. Overall, our research highlights the potential of XAI algorithms in feature selection, significantly reducing the dimensionality of HSI data without compromising performance. This approach establishes a methodological foundation for designing a cost-effective multispectral imaging system capable of distinguishing the botanical origin of honey.